K-space and image domain collaborative energy-based model for parallel MRI reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Signal Processing

Background:

• Magnetic Resonance (MR) imaging is crucial for medical diagnostics but suffers from long acquisition times.
• Deep learning models have been explored to accelerate MR imaging, but often lack direct k-space integration.
• Deep generative models offer promise for robustness and flexibility in image reconstruction.

Purpose of the Study:

• To develop a novel deep generative model for accelerated MR image reconstruction.
• To enable direct learning and application in the k-space domain.
• To investigate the performance of deep generative models in hybrid domains.

Main Methods:

• A collaborative generative model integrating k-space and image domains was proposed, utilizing deep energy-based models.
• The model was designed to estimate MR data from undersampled measurements.
• Parallel and sequential processing orders were implemented for enhanced performance.

Main Results:

• The proposed model demonstrated superior performance compared to state-of-the-art methods.
• Experimental results showed reduced error in reconstruction accuracy.
• The model exhibited increased stability across various acceleration factors.

Conclusions:

• The developed k-space and image domain collaborative generative model effectively accelerates MR imaging.
• This approach offers improved accuracy and stability for MR data reconstruction from undersampled measurements.
• The findings contribute to making MR examinations more accessible through faster acquisition times.